[PATCH] D137525: AMDGPU: Drop and upgrade llvm.amdgcn.atomic.inc/dec to atomicrmw

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/GlobalISel/atomicrmw_uinc_wrap.ll:70
 ; GFX10-NEXT:    s_load_dwordx2 s[0:1], s[4:5], 0x0
+; GFX10-NEXT:    s_waitcnt vmcnt(0) lgkmcnt(0)
+; GFX10-NEXT:    s_waitcnt_vscnt null, 0x0
----------------
cdevadas wrote:
> There are additional waitcnt instructions in most of the lit pattern changes. Were they regressions earlier?
Without an ordering passed through to the memory legalizer, these weren’t actually treated as atomics. They were not properly legalized before 


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D137525/new/

https://reviews.llvm.org/D137525